141B Economics Economics of Government Behavior II
Professor Francesca Mazzolari Winter 2008 Lecture 2
Today's plan
Review of Microeconomic concepts (conclude) Empirical tools of Applied Microeconomics Social Insurance (start)
Empirical tools of Public Finance
Correlation vs causation Randomized trials Observational data
Time-series analysis Cross-sectional analysis Quasi-experiments ("Natural" experiments)]
Empirical Public Finance
Using data and statistical methodologies to measure the impact of government policy on individuals and markets Key issue: separating causation from correlation.
Correlated means that two economic variables move together Causal means that one of the variables is causing the movement in the other
Critical for government policy to understand the difference
For policy purposes, what we care about is causation
Methods that economists rely on to learn about the causal effects of government policy
Randomized trials Analysis of observational data
Examples: causation and correlation
Example 1: When there are a lot of people at the bus stop, the bus tends to arrive soon Example 2: Students get higher SAT scores if they have teachers with more years of experience
Easy to imagine a causal channel
If experience makes you a better teacher, your students could quite plausibly get higher scores
But there's also correlation
teachers with more experience get to pick which school they teach at within the district, and they might choose schools where students have higher scores
Correlation versus causation: health insurance (ex. 3)
Wealth Risk aversion
Health outcome Health Insurance
The Problem
For any correlation between two variables A and B, there are three possible explanations:
A is causing B B is causing A Some other factor is causing both
Correlation versus causation: SAT example (ex. 4)
SAT preparation courses
In 1988, Harvard interviewed its freshmen and found those who took SAT "coaching" courses scored 63 points lower than those who did not One dean concluded that the SAT courses were unhelpful and "the coaching industry is playing on parental anxiety" The possibilities could be: SAT prep courses worsen preparation for the SATs Those with poorer test taking ability take prep courses to try to catch up Those who are generally nervous both like to take prep courses and do the worst on standardized exams
Random assignment
Think of medical model: randomly assign subjects to
Treatment group, which gets the medical treatment Control group, which does not
Project STAR in Tennessee
Treatment was small class (13-17), control was standard class (22-25) Randomly assign kids to small or standard classes
Both treatments within each school
Finding: smaller classes did help
Simple experimental set-up
Group (randomly assigned) Treatment (class size=13) Control (class size=25) Treatment effect Test score A B A-B
The Problem of Bias
With random assignment, the assignment of the "intervention" is not determined by anything about the subjects In the SAT example, the "treatment" group members are those who took the coaching course; the "control" group members are those who did not.
The assignment of the intervention was not random This means the treatment and control groups are not identical
The Problem of Bias
Bias represents any source of difference between treatment and control groups that is correlated with the treatment, but not due to the treatment By definition, such differences do not exist in a randomized trial, since the groups are not different in any consistent fashion As a result, randomized trials have no bias
they are the "gold standard" for empirically estimating causal effects
The Need to Go Beyond Randomized Trials
Randomized trials present some problems:
They can be expensive They can take a long time to complete They may raise ethical issues The inferences from them may not generalize to the population as a whole
Individuals may behave differently when they know they are being watched
Subjects may drop out of the experiment for non-random reasons, a problem known as attrition
Observational data
More often researchers use observational data, data generated from individual behavior observed in the real world There are 4 approaches researchers use to estimate causal effects with observational data:
1. 2. 3. 4. Time series analysis Cross-sectional regression analysis Quasi-experiments (Structural modeling)
1. Time Series Analysis
Time series analysis documents the correlation between the variables of interest over time Example:
Temporary Assistance to Needy Families (TANF)
program that provides cash benefits to single mothers whose income is below a specified level
Question: by only providing benefits to very low income single mothers, is TANF discouraging them to work? Data: time series of the TANF income guarantee and the labor supply of single mothers over time
Figure 3.1
Time Series Analysis
Time series analysis might not be all that useful But if there are sharp changes in a policy variable over time, then there may be some room for valid inference Example:
Question: do higher prices of cigarettes reduce youth smoking rates? Policy changes:
Cigarette price war in April 1993 Tobacco settlement in 1998
Figure 3.2
2. Cross-Sectional Regression Analysis
"Cross-sectional" means comparing many individuals at one point in time Regression analysis
It describes the relationship between the dependent variable (e.g.: labor supply), and the independent variables (e.g.: TANF benefits).
Hoursi = " + # TANFi + $i
!
Hours of work per year 1,800 1,500 1,200 900 600 300 0
$0
The figure graphs labor supply (vertical axis) against dollars of TANF benefits (horizontal axis) negative relationship Linear regression line: best linear approximation to the set of points Note that the line does not fit the points perfectly Slope: -110 : doubling TANF reduces work by 110 hours
$1-99 $100- $250- $500249 499 999
$1,0002,499
2,5004,999
TANF benefits received
Figure 3.4 CPS Data
Cross-Sectional Regression Analysis
This line corresponds to the regression:
Hoursi = " + # TANFi + $i
one observation for each mother "i" is the constant term is the slope coefficient is the error term that represents the difference for each observation between its actual value and its predicted value on the regression line
!
Cross-Sectional Regression Analysis
As in the time-series analysis, interpreting the results is potentially problematic
One interpretation is that higher TANF benefits "cause" lower labor supply Another interpretation is that single mothers with a greater "taste" for leisure (or higher barriers to work) get higher TANF benefits due to the program benefit calculation.
Add controls
Adding control variables changes the regression to:
Hoursi = " + # TANFi + $ Controli + %i
For example: race, education, age, and location
Including these controls reduce the systematic differences between different groups
But it is unlikely that control variables will ever completely solve the problem
There is no control variable for "taste for leisure"! There might be barriers to work faced by welfare recipients that we cannot observe and measure
3. Quasi-Experiments
Economists typically cannot set up randomized trials for many public policy discussions.
Yet, the time-series and cross-sectional approaches are often unsatisfactory
Quasi-experiments are changes in the economic environment that create roughly identical treatment and control groups for studying the effect of that environmental change
This allows researchers to take advantage of randomization created by external forces
TANF Example: state-time variation
Suppose, for example, that
Arkansas cut its TANF benefit by 20% in 1997 we have a large sample of single mothers in Arkansas in 1996 and 1998 By examining hours of work in Arkansas, we obtain:
HOURSAR,1998 - HOURSAR,1996
Arkansas
1996 Benefit Guarantee Hours of Work Per Year $5,000 1,000 1998 $4,000 1,200 Difference -$1,000 200
TANF example
If we studied single mothers in Arkansas alone
single mothers in 1996 are the control group, and those in 1998 are the treatment group
Potential criticisms: the national economy was growing exceptionally fast during this period Include the extra step of comparing the treatment group for whom the policy changed to a control group whom for it did not
For ex., let's say Lousiana did not raise benefits in the same period
TANF Example
Quasi-experimental design:
Women in Arkansas are the treatment group Women in Louisiana are the control
Compute the 96-98 change in labor supply in each state
HOURSAR,1998 - HOURSAR,1996 HOURSLA,1998 - HOURSLA,1996
Then examine the difference between treatment (Arkansas) and control (Louisiana)
Table 3.1
Using Quasi-Experimental Variation Arkansas
1996 Benefit Guarantee Hours of Work Per Year $5,000 1,000 1998 $4,000 1,200 1998 $5,000 1,100 Difference -$1,000 200 Difference $0 50
Louisiana
1996 Benefit Guarantee Hours of Work Per Year $5,000 1,050
The difference-in-difference estimator
DD estimator
the difference between the changes in outcomes for the treatment group that experiences an intervention and a control group that does not
The DD estimator is:
(HOURS
AK ,1998
! HOURS AK ,1996 )! (HOURS LA,1998 ! HOURS LA,1996 )
True experimental set-up (ex: STAR in Tennessee)
Group (randomly assigned) Treatment (class size=13) Control (class size=25) Treatment effect Test score A B A-B
Difference-in-differences set-up (ex: use a policy change)
Treatment POST PRE A C Control B D B-D Treatmentcontrol A-B C-D
(A-B)-(C-D) =(A-C)-(B-D) =DD est.
POST-PRE A-C
Problems with quasi-experimental analysis
This approach also has problems, however
It is possible that the economic boom affected Arkansas differently than it did Louisiana More generally, single mothers may be different across states Policy endogeneity!
What to do with non-experimental data?
Control for other variables Look for "natural" sources of variation Be a critical consumer of research!
Always ask what determines "treatment" and if that determinant is correlated with the outcome
Social Insurance: Outline
I. What is social insurance? II. Why have social insurance? III. Value of insurance IV. Adverse selection in insurance markets V. Role of social insurance VI. Consumption-smoothing benefits VII. Moral hazard costs VIII. Financing social insuranceGruber Ch 20, pp.600-5
Gruber Ch. 12
I. What is social insurance?
What is insurance?
Examples
health insurance, automobile insurance, life insurance, and casualty and property insurance
Common Structure
Individuals pay money to an insurer (called an insurance premium) In return, the insurer promises to make some payment to the insured party if an adverse event occurs
Private market failures
I.
Information asymmetry Example: market of used cars (Akerlof, 1970) In insurance mkts: the purchasers of insurance know more about their insurable risks than the insurer does Private markets for risk may be limited or missing due to adverse selection Government can improve efficiency by providing public insurance programs But government intervention can also introduce new costs due to moral hazard
Key social insurance programs I.
1.
Unemployment insurance
job loss
2.
Workers' Compensation
on-the-job injury
3.
Disability Insurance
career-ending disability
4.
Social Security
earnings loss due to retirement or death
5.
Medicare
out-of-pocket health care expenses in old age
Common features
1. Participation is mandatory 2. Eligibility and benefits depend on contributions 3. Benefits are not means-tested 4. Benefit receipt is tied to the occurrence of an event
I.
Figure 12.1
I.
Other, 22.0% Defense, 18.8%
Breakdown of Federal Government Spending
Other, 21.6%
Income Security, 5.0% Social Security, 3.6% Health, 0.4%
Income Security, 15.5%
Defense, 69.4%
Health, 21.7%
1953
Social Security, 22.0%
2003
II. Why have social insurance?
1. Externalities
Physical Fiscal/financial Private health insurance: 12% of claims paid Medicare: 2% of claims paid Individuals are myopic Or they understate probabilities of negative events
2. Administrative costs
3. Paternalism
Why have social insurance? (cont.)
4. Redistribution
Ex post toward those with bad luck Benefit formulas are progressive Technologies removing problems of asymmetric information might worsen the redistribution problem
II.
5. Adverse selection
Asymmetric information: buyers have private information about their "type" Can cause markets to operate inefficiently, or even not at all
Adverse selection
II.
Asymmetric information: individuals seeking insurance know more about their own risk level than does the insurer Those most likely to have the adverse outcome have higher odds to select insurance
Health insurance example
II.
Imagine expected annual health care expenditures are uniformly distributed between $0 and $1000
$0
$500
$1000
Health insurance example
II.
Who buys insurance if the price is $500?
$0
$400 $500
$700
$1000
Health insurance example
II.
Who buys insurance if the price is $700?
$0
$400 $500
$600 $700 $800
$1000
III. Value of insurance
Model of the decision to purchase insurance
Why is insurance valued by consumers?
Insurance is valuable to people because of the principle of diminishing marginal utility
Individuals prefer 2 years of average consumption (C) Excessiv...